|Publication number||US8027527 B2|
|Application number||US 11/472,368|
|Publication date||Sep 27, 2011|
|Filing date||Jun 22, 2006|
|Priority date||Apr 10, 2001|
|Also published as||US20060238755, US20060239536|
|Publication number||11472368, 472368, US 8027527 B2, US 8027527B2, US-B2-8027527, US8027527 B2, US8027527B2|
|Inventors||Hisae Shibuya, Yuji Takagi|
|Original Assignee||Hitachi, Ltd., Hitachi High-Technologies Corporation|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (9), Referenced by (3), Classifications (13), Legal Events (1)|
|External Links: USPTO, USPTO Assignment, Espacenet|
This application is the first of two concurrently filed continuation Applications of U.S. Ser. No. 11,095,614, filed Apr. 1, 2005 now U.S. Pat. No. 7,084,968, which is a continuation of U.S. Ser. No. 10/119,018, filed Apr. 10, 2002 (now U.S. Pat. No. 6,876,445), the contents of which are incorporated herein by reference.
The present invention relates to a defect data analyzing method for analyzing defect distribution states from the defect data detected by inspection apparatus in a semiconductor device manufacturing process in which circuit patterns are formed on a semiconductor substrate.
In the semiconductor device manufacturing process in which circuit patterns are formed on a semiconductor substrate (hereafter, referred to as semiconductor substrate production process), pattern defect inspection or foreign matter inspection is executed after each process, and the inspection results are analyzed, in order to improve and stabilize the yield. In other words, the operator observes the detected defects on an optical microscope or scanning electron microscope to know the kinds of the defects and identify the causes of the defects. This operation is called review. A method for making effective review is disclosed in JP-A-10-214866. In this method, the region in which defects are concentrated is recognized as a cluster from the defect distribution, and a review point is selected on the basis of the area and shape of the cluster.
Another method for analyzing the inspection result is proposed to try to estimate the cause of defects in apparatus or process from the analysis of defect distribution states. JP-A-6-61314 describes that wafers are grouped according to the state in which the defect map has clusters, and it is decided if they have similarities to known patterns, thereby identifying the cause of defects. In addition, U.S. Pat. No. 5,982,920 describes that each defect is classified as one of minute cluster, linear cluster, indefinite-form cluster and global type other than cluster, and related to a cause of defect.
It is an object of the invention to provide a defect data analyzing method capable of extracting a relatively high-density region even if the detected defects have a low-density distribution.
It is another object of the invention to provide a defect data analyzing method capable of fast classifying clusters of high-density regions of detected defects at a practical level.
The method described in the above-given JP-A-10-214866 needs to understand the cluster and then specify the shape. Therefore, it cannot be used for the distribution that is so thin that the cluster cannot be recognized. The method disclosed in the above JP-A-6-61314 or U.S. Pat. No. 5,982,920 also similarly requires to recognize the cluster, and thus cannot be used for thin defect distributions. However, in order to early detect an abnormal state of apparatus or process, it is necessary to be capable of treating such thin defect distributions.
In addition, according to the method described in the above JP-A-6-61314, patterns of an indeterminate form are registered because known patterns to be used for comparison are produced on the basis of actual wafers. The method in the U.S. Pat. No. 5,982,920 makes clusters of various forms of lines or of indeterminate forms be classified mixed as one type. In either case, when the cause of defect is specified by manpower, skill is required. For automatic inspection, an enormous amount of data must be accumulated, and as a result the computation time increases. Therefore, information that makes the defect cause identification by manpower easy is required to be fast produced.
In order to achieve the above object, according to the invention, there is provided a defect data analyzing method for classifying defects, on the basis of the position coordinates of defects detected by an inspection apparatus, into at least one kind of distribution feature category, or any one of repetitive defect, congestion defect, linear distribution defect, ring/lump distribution defect and random defect.
According to the invention, since a high-density region image representing a high-density portion of defects is generated on the basis of the position coordinates of defects detected by an inspection apparatus, the high-density region is not required to recognize as cluster, and a relatively high-density region can be extracted even if the distribution is thin. Moreover, according to the invention, since the high-density region image is classified into any one of a plurality of previously registered geometric patterns, the defect distribution states can be analyzed at a practical level in both computing time and storage capacity. Since meaning can be previously created in each of the plurality of geometric patterns, the cause of defect in apparatus or process can be identified with ease.
Other objects, features and advantages of the invention will become apparent from the following description of the embodiments of the invention taken in conjunction with the accompanying drawings.
Embodiments of the invention will be described with reference to the accompanying drawings.
In the first embodiment of the defect data analyzing method according to the invention, defect data 102 produced from a semiconductor substrate inspection apparatus 101 includes at least the coordinates of defect position. Shown at 103 a, 103 b are wafer maps for showing the positions of defects expressed by a coordinate system with its origin selected as one point from the wafer. The coordinates of defect that the inspection apparatus produces can be expressed by a coordinate system with its origin selected as one point from the wafer or by coordinate systems with their origins selected from the respective chips of the wafer. In the former case, the coordinate values X, Y of defect are used as they are, but in the latter case it is necessary that the coordinates of defect within a chip be converted to be on another coordinate system by using chip arrangement information and chip size information. The defects on the wafer map are respectively classified on the basis of its defect distribution into distribution characteristic categories by the discrimination processing of repetitive defect, congestive defect, linear distribution, ring/line-shaped distribution and extraction processing of random defect.
The repetitive defect occurs over a plurality of chips to be at substantially the same positions within each chip. Shown at 104 a, 104 b is the repetitive defect as indicated by dots corresponding to those in the maps 103 a, 103 b.
The congestion defect has a very small distance to the adjacent defect on the wafer map, and normally called cluster defect. Shown at 105 a, 105 b are congestion defects that correspond to those of the maps 103 a, 103 b as indicated by dots. For congestion defect discrimination, defects with the distance equal to or less than a predetermined threshold are connected to form a group, and the group having a predetermined number of defects is detected as congestion defect. A closest point Voronoi diagram corresponding to the wafer map is generated to determine the distances between the adjacent defects, and it is checked if the distance to the adjacent defects is equal to or less than a threshold. Alternatively, the wafer map is partitioned into lattice areas, and it is previously examined which lattice each point belongs to. It is possible to check if the distance between the defects in the same lattice or in the adjacent lattices is equal to or less than a threshold. The congestion defect has its group number attached for each group, and the gravitation center coordinates, maximum x/y coordinates, minimum x/y coordinates, area, defect density and constituent defect number that characterize the group are calculated and recorded.
The linear distribution defect has a linear distribution of high-density defects. The portion surrounded by a square in 106 a, 106 b shows the linear distribution defect corresponding to that of 103 a, 103 b. The linear distribution does not strictly show a straight line, but has a certain width.
The ring/lump distribution is a distribution having a ring-shaped or lump-shaped distribution of high-density defects. The portions surrounded by the solid line in 107 a, 107 b show the ring/lump distribution defect corresponding to those in 103 a, 103 b.
The other type of detect than the above types of defect is extracted as random defect. Shown at 108 a, 108 b are the random defects that correspond to those in 103 a, 103 b.
The linear distribution defect identifying method according to the invention will be described in detail with reference to
Hough transform is often used for linear detection.
Hough transform is also used for the identification of linear distribution defect according to the invention.
Then, the width W in ρ-direction is determined of the portions of α% of the maximum ballot number of the proposed straight line as shown in
Then, a proper method such as discrimination analysis method is used to binarize the dark and light image as shown in
The ring/lump distribution defect identifying method of the invention will be described in detail.
Each step will be described in detail.
A description will be first made of a method for producing the high density region image 704 on the basis of the defect position coordinates.
First, a closest point Voronoi diagram for all defect coordinates is generated, and the area of the Voronoi region at each point is determined (S801). If this calculation is made in the above linear distribution defect discrimination processing, the calculation result may be used. Then, the histogram of the area of the Voronoi region is calculated (S802). In this case, the defects classified as any type of repetitive defect, congestion defect and linear distribution defect are not included in the calculation of histogram. The highest frequency value is determined from the histogram, and multiplied by a predetermined number (assumed as N), and the square root of the multiplication result is calculated and used as the image size (S803). The image size is determined so that the whole wafer can be fallen just within a single image, and a dark and light image is generated with the number of defects per pixel being expressed by tint (S804). This image is generated by incrementing the pixel value at the coordinates of each defect with all pixels initially set to 0. At this time, when N is too large, the image size becomes small, making it difficult to discriminate shapes, and thus N should be selected to be about 5. If N is too small contrary to the above case, the thickness difference between the high-density and low-density portions becomes small, making it difficult to discriminate. Next, the image is binarized by use of a predetermined threshold (represented by T) to be a binary image of 1's indicating high density portion and 0's indicating low density portion (S805). Here, T and N should be made substantially equal. The obtained image has a pattern of a relatively high defect density. Finally this pattern undergoes expansion and contraction processing to form the high-density region image 704 (S806). Thus, by generating the image using this method, a relatively high density region can be extracted with a proper image size set even if the image has a thin distribution. In addition, another method can be considered in which the dark and light image is generated in step S804 and binarized in step S805. Each defect is weighted in proportion to the shortest distance to the adjacent defect, the square of the shortest distance or the Voronoi region area, and added to the pixels at the corresponding positions, thereby generating the dark and light image. This image is binarized by use of a discrimination analysis method.
A description will be made of the method for automatically producing the geometric dictionary pattern images 705.
The geometric dictionary pattern images 705 are automatically generated with the ring pattern and lump pattern separated. The image size is determined as with that of high-density region image 704. The pattern is generated by using the largest circle to be depicted on the image as a reference according to the following method.
The ring patterns are generated by dividing the reference circle equally in the radius direction to be concentric circles, and combining small regions resulting from equal fan-shape division of the circles toward the central angle as shown in
The circular rings are not always required to divide as in the radius direction and central angle direction. Also, they are not necessary to divide equally. However, equal division is desirable not to prevent the processing from being complicated.
The lump-shaped patterns are generated by dividing the reference circle equally in the horizontal direction by elliptical shapes with the long axis selected as the vertical diameter, dividing equally in the vertical direction by elliptical shapes with the long axis selected as the horizontal diameter and combining the resulting small regions.
By automatically generating the geometrical dictionary pattern images 705 in the above way it is possible to suppress the pattern number to a practical level, and thus achieve fast processing.
A description will be made of a method for calculating the degree of coincidence between the high-density region image 704 and each of dictionary pattern images 705 with reference to
In the above method, the pattern 706 is detected without fail, but in practice, it should be decided that any pattern is not produced at all when the density is very low or when the density difference is small. Therefore, the detected pattern 706 is verified so that the presence and absence of pattern is decided. A description will be made of a method for deciding the presence or absence of pattern and a method for adjusting the sensitivity of pattern detection. In order to decide the presence or absence of pattern, it is necessary that a dark and light image be generated by a weighting method and used when the high-density region image 704 is produced. The detected pattern image 706 and the dark and light image are superimposed on each other, and the ratio C of decision analysis values of the inside and outside of pattern is calculated from the pixel values of the dark and light image according to the following equation.
The ratio C is compared with a predetermined threshold. If it is equal to or smaller than the threshold, it is decided that there is no pattern. In addition, if the highest frequency is lower than a predetermined value in the histogram obtained in step S802, it is decided that there is no pattern without making the following processing. If the former threshold is selected to be small, a pattern with a smaller density difference can be detected. If the latter threshold is set to be small, a pattern with a smaller defect density can be detected. Therefore, if these thresholds are specified by the user, the user can adjust the sensitivity.
Incidentally, the semiconductor substrate is not always examined for all surface because of the restriction to the throughput, but it is often partially examined by specifying inspection or no inspection for each row of chips. In this case, it is difficult for the pattern detection to be made by the above ring/lump discrimination method. Another method for discriminating ring/lump distributions with the above problem solved will be described with reference to
The user pattern registration method will be described below.
The user pattern registration method may include both types of method or either one. Also, it may be different from the above one.
In the second embodiment of the defect data analysis method according to the invention, information associated with the automatically generated geometrical dictionary pattern 705 and user's registered pattern 707 is added in the ring/lump distribution defect discrimination processing, and the pattern image 706 is selected according to the same method as in the first embodiment.
The information to be added includes pattern position, size, shape, degree of importance, and specified items. The pattern position, size and shape of the geometrical dictionary pattern 705 are roughly classified according to a constant rule, and default information is added. The positions of the lump-shaped pattern are grouped into upper right, top, upper left, right, center, left, lower right, bottom and lower left. The sizes thereof are classed as large, medium and small, and the shapes are lump. For example,
Moreover, means is provided for manually grouping the geometrical dictionary pattern 705 and user's registered pattern and collectively adding information, and the same information can be added to the pattern that the user decides to be the same. Accordingly, the resolution of position and size information can be adjusted. The degree-of-importance information is entered by the user. The specific items to be entered include information directly associated with cause of defect.
The first embodiment of the inspection apparatus having the defect data analysis method according to the invention will be described. Hereafter, repetitive defect, congestion defect, linear distribution defect, ring/lump distribution defect and random defect are called distribution feature category. Known inspection apparatus for semiconductor substrate include foreign substance inspection apparatus, optical type pattern defect inspection apparatus and SEM type pattern defect inspection apparatus. The inspection apparatus according to the invention makes inspection for semiconductor substrate by the same known method as either one of these inspection apparatus, classifies the obtained defect data into the distribution feature category by the above method and produces the category information together with the defect data information.
Pattern information adding means 301 adds pattern-related information such as pattern position, size, shape, degree of importance and specified item to a plurality of automatically generated geometrical dictionary patterns and user's registered dictionary patterns. It has a portion to add default information to the geometrical dictionary pattern, and a portion to manually add information to the geometrical dictionary pattern and user's registered dictionary pattern. The addition of pattern information is made offline. The result is at least once stored as a pattern information file 311 in storage means 302 such as hard disk.
The defect data analysis can be executed in both in-line mode and offline mode.
Inspection means 303 makes wafer inspection by a well known method, and causes defect data 102 including at least defect position coordinates to be stored in the storage means 302.
In in-line mode, the inspection means 303 sends notice of inspection completion to defect data analysis means 304 when inspection of each wafer is finished. The defect data analysis means 304 reads in defect data 102 of each wafer from the storage means 302. Alternatively, it may be so constructed as to transmit and receive defect data 102 without the intervention of storage means 302.
In offline mode, the defect data analysis means 304 is ordered to read in defect data 102 of wafer by the operator.
The defect data analysis means 304 classifies defect data into distribution feature category on the basis of defect position coordinates, and adds distribution feature category number and intra-category group number to defect data of each defect. When a ring/lump distribution defect is detected, it causes the pattern information file 311 to be read from the storage means 302, adds pattern image number or pattern information to the defect data of the ring/lump distribution defect, and makes it be stored in the storage means 302 as defect data 102 including the results of both inspection and analysis. In in-line mode, the inspection means 303 is informed of analysis completion.
The result of classification to distribution feature category is displayed on display means 305. It may be displayed on a wafer map with a different color for each category, or a wafer map may be generated for each category and displayed as shown in
User pattern registration means 306 generates user's registered patterns according to the order by the operator, and causes pattern code or pattern image to be stored in the storage means 302, depending on the difference in the above method. It is also possible to construct with the user pattern registration means 306 not provided.
A description will be made of the second embodiment of the inspection apparatus having the defect data analysis method according to the invention. The second embodiment inspects the semiconductor substrate by a well known method, makes the classification to distribution feature category on the basis of the obtained defect data information by the above method, makes sampling on the basis of a specified rule for each category, and produces the classification result and sampling result together with defect data.
The pattern information adding means 301 adds sampling condition number in addition to the above information. It manually groups the dictionary patterns using the same sampling condition, and attaches the same sampling number. When there is no sampling condition corresponding to the number, sampling condition generation means 307 generates the sampling condition.
The sampling condition generation means 307 generates the sampling condition for determining a defect for review. It manually generates the sampling conditions associated with the patterns of ring/lump distribution defects, and sampling conditions at each of the other distribution feature categories, and makes them be stored as a sampling condition file 312 in the storage means 302. The sampling conditions are, for example, sampling number, sampling rate and sampling method. The values of sampling number and sampling rate are entered after selecting either one for the designation. The sampling number is determined by the sampling rate to the number of defects within a pattern. Sampling manners such as random and defect number order are listed so that the user can select them. When defect number order is specified, sampling is made at intervals matched with the sampling rate.
The operations of the inspection means 303 and defect data analysis means 304 are the same as in the first embodiment.
Sampling means 308, after completion of defect analysis, reads in the sampling condition 312 and the single-wafer defect data 102 with distribution feature information added from the storage means 302. The transmission and reception of defect data 102 may be made without the intervention of the storage means 302. The condition of sampling is determined by the distribution feature category of defect data. If the category is ring/lump distribution defect, sampling is made by use of the condition associated with the added pattern information. If the category is the other one, sampling is made by use of the condition of each distribution feature category. As the result of the sampling, a flag of review or not is added to the defect data, and the defect data 102 including the results of inspection, analysis and sampling is stored in the storage means 302.
The pattern image 706 selected by the defect data analysis means 304 is displayed on the display means 305 together with information associated with the wafer to be inspected, and the pattern information added to pattern. The defect sampled on the wafer map is displayed in a different color or symbol from the other defects.
The user pattern registration means 306 is operated in the same way as in the first embodiment, but may be omitted.
A description will be made of a review system having the defect data analysis method according to the invention.
On a network are connected a foreign substance inspection apparatus 401, an optical type pattern defect inspection apparatus 402, an SEM type pattern defect inspection apparatus 403, a defect data analysis apparatus 404 and a review apparatus 405. Each inspection apparatus 401-403 and the review apparatus 405 are installed within a clean room. The defect data analysis apparatus 404 may be installed anywhere.
The results from the inspection apparatus 401˜403 are produced as a file of defect data 102 of the same format, and transferred to the defect data analysis apparatus 404. The defect data analysis apparatus 404 is constructed as in
The review apparatus 405 reads in the defect data 102 with the sampling flag added of the wafer to be reviewed, reviews the defect according to the information and manually or automatically classifies the defect. As a result of classification, a category number is added to the defect data 102, and the defect data is produced as a file and transferred to the defect data analysis apparatus 405. At the same time, the related review image is also transferred. The defect data analysis apparatus 405 generates a report and stores it on the basis of the defect distribution pattern of the wafer to be analyzed, and the defect review result. The content of the report includes information of wafer to be inspected, the result of classification to distribution feature category, and review image. Since distribution feature and defect review image are simultaneously displayed, the cause can be easily presumed. According to user's request, the detailed information such as estimated rate of defect types or other analyzed results by a well known analysis means may be given which are calculated from the sampling position information, review result and sampling rate. Any format may be used for the report, but HTML format is used. The report is uploaded to a data server 406 connected via the Internet or intranet so that it can be read from an arbitrary terminal 407 that can be connected to the server.
While an example of three kinds of inspection apparatus and one review apparatus connected was shown above, at least one inspection apparatus and one review apparatus may be connected. Moreover, a plurality of various inspection apparatus and review apparatus may be provided. If each inspection apparatus has the same construction as in the second embodiment of the inspection apparatus according to the invention, the function of defect data analyzing apparatus can be achieved by any one of the inspection apparatus, and thus it is not necessary to separately provide the defect data analyzing apparatus.
While the above defect data analyzing method, and the inspection apparatus or review system having this method are provided to fundamentally process defect data of a single wafer, a plurality of wafers can be analyzed together as long as the defect distribution discrimination but not sampling is tried to make. In this case, first the defect position coordinates of a plurality of wafers are superimposed, and a high-density region image is generated on the basis of the superimposed defect position coordinates.
It should be further understood by those skilled in the art that the foregoing description has been made on embodiments of the invention and that various changes and modifications may be made in the invention without departing from the spirit of the invention and the scope of the appended claims.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US5240866 *||Feb 3, 1992||Aug 31, 1993||At&T Bell Laboratories||Method for characterizing failed circuits on semiconductor wafers|
|US5479252||Jun 17, 1993||Dec 26, 1995||Ultrapointe Corporation||Laser imaging system for inspection and analysis of sub-micron particles|
|US5831865||Jul 14, 1997||Nov 3, 1998||Advanced Micro Devices, Inc.||Method and system for declusturing semiconductor defect data|
|US5943437 *||Oct 7, 1996||Aug 24, 1999||Kabushiki Kaisha Kobe Seiko Sho||Method and apparatus for classifying a defect on a semiconductor wafer|
|US5982920||Jan 8, 1997||Nov 9, 1999||Lockheed Martin Energy Research Corp. Oak Ridge National Laboratory||Automated defect spatial signature analysis for semiconductor manufacturing process|
|US6097887||Oct 27, 1997||Aug 1, 2000||Kla-Tencor Corporation||Software system and method for graphically building customized recipe flowcharts|
|US6317859||Jun 9, 1999||Nov 13, 2001||International Business Machines Corporation||Method and system for determining critical area for circuit layouts|
|JPH0661314A||Title not available|
|JPH10214866A||Title not available|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US8824773 *||Nov 19, 2010||Sep 2, 2014||Hitachi High-Technologies Corporation||Defect observation method and defect observation device|
|US8890084 *||Sep 3, 2013||Nov 18, 2014||United Microelectronics Corp.||Method for analyzing circuit pattern|
|US20130140457 *||Nov 19, 2010||Jun 6, 2013||Yohei Minekawa||Defect observation method and defect observation device|
|International Classification||G06K9/00, G06T7/00, G01N21/95|
|Cooperative Classification||G01N21/956, G06T7/0006, G06T7/001, G06T2207/30148, G01N21/9501|
|European Classification||G06T7/00B1R, G01N21/95A, G06T7/00B1D, G01N21/956|